An image-processing ranging system using a pair of cameras typically utilizes a range-image method. In this method, the amount of processing is large, and a high-performance CPU and a large-scale circuit are required. In order to reduce the amount of processing and improve the accuracy for distance determination, portions having many features are extracted from an image, and distance determination is performed only for the positions where the features are extracted. Feature extraction is performed by, for example, an edge extracting process, etc. Distance determination is performed by a pattern matching process. Thus, high performance distance determination, at the same level as the above range-image system, can be carried out while the amount of processing is reduced.
However, the problems listed below occur when edge extracting and pattern matching processes are used.
(1) If a vertical edge is not extracted in a feature extracting process, the distance to an object to be recognized in front of a ranging device cannot be determined. In particular, for an object having many oblique edges, portions having many features are difficult to detect by an edge extracting process.
(2) Accurate feature extraction increases the required amount of processing, thereby increasing the time of processing. For this reason, the amount of processing must be reduced.
(3) The image of a windshield wiper or the like passing in front of a pair of cameras may be captured by the cameras. In such cases, it must be prevented that the distance is not determined or determined improperly due to the image of the windshield wiper or the like.